Queer Coded: Supervised ML to Predict LGBTQ+ Acceptance in American Neighborhoods

Final Project Progress Memo 1– STAT 301-3 Data Science 3 with R

Author

Vlad Nevirkovets, Sean Pascoe, Dylan Yan

Published

April 30, 2023

Github Repo Link

https://github.com/STAT301-3-2023SP/final-project-central-limit-heros

Prediction Project Outline

Introduction

Neighborhoods with a high queer population, or Gayborhoods, have been urban areas in which LGBTQ+ people have found community and built culture worldwide. However, these geographic areas serve as much more than the sexuality of their constituents, and have been cited as yielding robust creative economies, as well as a welcoming environment for those of many identities. Knowing this, the identity of a certain locale as a Gayborhood becomes a crucial sociological metric, with neighborhoods with a more prevalent queer identity driving social liberalism in the face of prejudice. Our analysis thus focuses on building a location-based model that can uses a variety of parameters including housing, land use, and non-queer demographic data to predict Gayborhood status, and, in so doing, determine whether an area is suitable for queer folks with the goal of advancing understanding of liberal areas.

Outcome Variable

In order to rank Gayborhood status, we are using a published dataset from data world1 that was analyzed to display distribution of queer communities in Jan Diehm’s 2018 article Men are from Chelsea, Women are from Park Slope. Specifically, the metrics in this dataset are combined to form a gayborhood index, which we will use as the supervising variable for our machine learning project. This index is a wholistic measure of queerness, derived primarily from the following measures:

  • Same-sex joint tax filers
  • Unmarried partner same sex households
  • Number of Gay Bars
  • Whether or not a pride parade routes through the region

Regression - predicting index scores Include the link somewhere

Include density plot or histogram here, does it need to be transformed?

Predictor Variables

Because this project is based primarily on data by ZCTA, a US Census-defined zipcode proxy, we are able to pool data from three main sources, to encompass 3 major facets of urban life.

  1. Our first predictor set2 comes from the US Census, and provides a variety of information about housing characteristics including number of housing units in an area, rent prices, and indicators of development (phone reception, income to rent ratio, etc.). This is derived from the Census’s ACS 5-year survey estimates, as published in 2021.
  1. We also used US Census data to glean information about various demographic characteristics, as represented in our second data section3. This set primarily synthesizes demographic information that is not specifically related to sexuality (i.e. the parameters that went into the calculation of our outcome variable), but describe other characteristics of each area.
  1. We also opted to include data from Open ICPSR, which included information about parks4 in each geographical region, including number of open parks and proportion of ZCTA area that is park space. Since parks are public areas, they serve as a predictive metric for the culture of a community, and are thus important in describing neighborhood lifestyle.

Issues and Challenges

Choosing the right predictors, how to capture as much of the variation as possible?

Housing data is quite large, number of observations to number predictors is suspiciously close, need to narrow down the predictors before fitting models, census data also to some extent

Missing data, needs to be imputed which may take some time for the machine to run

Initial Steps

Resampling

How many folds and repeats? V-fold probably?

Recipes

Model Types

Comparison Metric

Probably RMSE?

Footnotes

  1. Jan Diehm. (2018). Gayborhoods [Data set]. The Pudding, data.world. https://data.world/the-pudding/gayborhoods. Accessed 10 April 2023.↩︎

  2. US Census. (2021). DP04: SELECTED HOUSING CHARACTERISTICS [Data set]. data.census.gov. https://data.census.gov/table?q=rent&g=010XX00US$8600000&tid=ACSDP5Y2021.DP04. Accessed 24 April 2023.↩︎

  3. US Census. (2015). DP05: ACS DEMOGRAPHIC AND HOUSING ESTIMATES [Data set]. data.census.gov. https://data.census.gov/table?g=010XX00US$8600000&tid=ACSDP5Y2015.DP05. Accessed 26 April 2023.↩︎

  4. Li, Mao, Melendez, Robert, Khan, Anam, Gomez-Lopez, Iris, Clarke, Philippa, and Chenoweth, Megan. (2020). National Neighborhood Data Archive (NaNDA): Parks by ZIP Code Tabulation Area, United States, 2018. Ann Arbor, MI: Inter-university Consortium for Political and Social Research. https://doi.org/10.3886/E119803V1. Accessed 23 April 2023↩︎